Every accountancy firm I speak to is having the same conversation. The partners know AI is reshaping professional services. A few people in the practice have experimented with ChatGPT. Maybe someone ran a lunch-and-learn. But nothing has changed how the firm actually produces work. The accounts still take the same number of hours. The management reports still get assembled the same way. The client advisory letters still get drafted from scratch each time.

The gap between "we should look at AI" and "AI has changed how we operate" feels wider than ever. And in most firms, nobody is quite sure whose job it is to close it.

Meanwhile, something significant is shifting in the background. A small number of professional services firms have stopped treating AI as a curiosity and started treating it as core infrastructure. Not a calculator that works faster. Not a chatbot that answers tax questions. A fundamental part of how the practice produces, reviews, and delivers work. The results are substantial enough that the rest of the profession needs to pay attention.

The Accountancy Productivity Problem

Accountancy has always had a tension at its core. The work that clients value most (advice, insight, strategic guidance) is also the work that gets squeezed out by the work that consumes the most hours (production, compliance, reporting, correspondence).

A typical engagement cycle tells the story. A partner wins the client relationship on the strength of their commercial acumen and industry knowledge. Then the delivery team spends the vast majority of their time on production: preparing accounts, compiling tax computations, drafting management reports, writing engagement letters, producing board packs, composing client correspondence, and managing the back-and-forth of information requests. The advisory conversations that the client actually values get compressed into whatever time is left.

Every firm has tried to improve this. Better practice management software. Standardised templates. Cloud bookkeeping integrations. Each improvement is incremental. The templates save formatting time. The cloud integrations reduce data entry. But the fundamental ratio between production hours and advisory hours has barely shifted in decades.

AI, implemented properly, shifts the ratio.

Why Generic AI Tools Have Not Worked

Most accountants who have tried AI have had the same experience. They opened ChatGPT, typed in a question about capital allowances or VAT treatment, got a response that was either too generic to be useful or confidently wrong on a technical detail, and closed the tab. That was their AI experiment.

The problem was never the underlying capability of the AI. The problem was that a general-purpose chatbot with no knowledge of your firm, your clients, your house style, or the specific regulatory framework you operate in will always produce generic output. It is the equivalent of asking a stranger off the street to draft a set of management accounts. They might have the technical knowledge, but they do not understand the context.

The current generation of AI (Claude in particular) works differently. It is not a pre-built accountancy tool. It is a general intelligence layer that you configure around your practice. You teach it your firm’s standards, your templates, your preferred formats, your analytical frameworks, and the specific regulatory and reporting requirements of your client base. Then it applies that understanding across everything it produces.

The difference is the gap between a calculator and a colleague who has worked at your firm for five years. The calculator processes numbers. The colleague understands your clients, knows your house style, and produces work that reflects how your firm thinks.

What “Running on Claude” Actually Means for an Accountancy Firm

A properly configured Claude implementation works in layers. Each one builds on the last, and together they transform what a practice can produce in a day.

Layer 1: Personalisation

Every user gets their own configuration. Not a shared login. A setup that reflects how they specifically work. Their writing style. Their preferred structure for client letters versus management reports versus board packs. The level of technical language they use with different client types. Claude learns these preferences and applies them consistently.

This is the detail that makes adoption stick. The reason most accountants abandon AI tools is that the outputs do not sound like them. A client letter drafted by a generic AI reads like it was written by a different firm. When Claude is configured to a specific person’s style, the output is a first draft that actually sounds like their work. The time between "AI draft" and "ready to send" drops from a rewrite to a quick review.

Layer 2: Shared Projects

A Project in Claude is a persistent workspace loaded with your firm’s content. Think of it as giving Claude access to the institutional knowledge that currently lives in partners’ heads, scattered across shared drives, or buried in last year’s working papers.

A client Project might contain: the prior year accounts, the engagement letter, the tax computations, previous management reports, notes from partner meetings, the client’s financial history, and any sector-specific regulatory requirements. When someone works within that Project, Claude draws on all of it automatically. The management report reflects the client’s specific circumstances. The draft accounts are consistent with prior year treatment. The advisory letter references the client’s actual position rather than starting from a blank page.

A technical Project might contain: your firm’s preferred treatment for common accounting issues, your house style guide, your standard disclosure templates, HMRC guidance notes for areas you deal with regularly, and examples of work that meets your quality standards. This is the difference between a tool that knows accounting and a tool that knows how your firm does accounting.

Layer 3: Skills

This is where the real leverage sits. A Skill is a reusable instruction set that encodes how your firm does a specific task. Not a template. A complete workflow that captures your quality standards, your preferred format, your analytical approach, and your firm’s voice.

Example: Management Accounts Commentary Skill

A client’s monthly figures are ready. The bookkeeper has reconciled the ledgers. Now someone needs to write the commentary. Traditionally this means an hour of a senior accountant’s time: pulling out the variances, explaining the movements, flagging the items the client needs to act on, and formatting it all into the firm’s standard report.

With the management accounts Skill, the accountant uploads the trial balance and invokes the Skill. Claude produces a structured commentary in your firm’s format: revenue and cost variances identified and explained, cash flow movements highlighted, key ratios calculated and benchmarked against prior periods, and actionable recommendations drafted for the client meeting. The accountant reviews, adjusts anything that needs their judgment, and sends. What took an hour takes fifteen minutes, and most of that is thinking, not typing.

Example: Year-End Accounts Review Skill

A set of draft accounts comes back from the preparation team. The review partner needs to check disclosure completeness, accounting policy consistency, Companies Act compliance, and presentation quality. The review Skill runs through the accounts systematically: flagging missing disclosures, checking that accounting policies match the treatments applied, identifying inconsistencies with prior year, and producing a structured list of review points with specific references.

The review that used to take a partner two hours of concentrated reading now starts with a structured analysis that highlights exactly where their attention is needed. The partner’s time goes to judgment calls, not line-by-line checking.

Example: Tax Advisory Letter Skill

A client calls with a question about structuring a property transaction. The partner knows the answer in broad terms but needs to produce a formal advice letter covering the CGT implications, SDLT considerations, potential incorporation relief, and the interaction with the client’s existing arrangements. The tax advisory Skill structures the analysis: identifies the relevant legislation, maps the client’s specific circumstances against the available reliefs, quantifies the tax impact under different scenarios, and produces a draft letter in the firm’s standard format with the partner’s personal tone. The letter goes out looking like it took a day. It took two hours, most of which was the partner thinking about the advice rather than producing the document.

Example: Client Onboarding Skill

A new client engagement needs an engagement letter, an information request, anti-money laundering documentation, and a preliminary scope of work. The onboarding Skill takes the basic client details and produces the complete pack: engagement letter tailored to the services being provided, a structured information request list specific to the client type, AML documentation requirements mapped to the risk profile, and a draft scope document. What used to take a mix of admin and senior time across a week is ready for review in minutes.

Each Skill gets refined over time. The management accounts Skill that works for a retail client gets adapted for a property portfolio. The tax advisory Skill gets tuned for different transaction types. The library grows, and the quality of every output improves because each iteration builds on what came before. After six months, your firm’s Skill library represents a genuine competitive asset: the accumulated expertise of the practice, encoded and reusable.

Layer 4: M365 Integration

Claude Team Plan connects natively to Microsoft 365: Outlook, SharePoint, OneDrive, Teams. This means Claude can read your existing documents, search your email correspondence, summarise client threads, and work with the files already in your systems. No copying data between tools. No switching between tabs. Claude operates inside the infrastructure you already use.

For an accountancy firm running on Microsoft (which is the vast majority), this removes the adoption barrier that kills most new tools: the friction of changing how people work.

Enjoyed this? Join the newsletter.

One email a week. What I'm building, learning, and what's actually working. No fluff.

The Economics

The cost structure is almost trivially small. A Claude Team Plan costs $20 per user per month. A firm of ten people costs roughly £1,900 per year. That is less than the annual cost of most practice management software add-ons. The return depends entirely on how well the tool is implemented.

Here is the calculation that matters for an accountancy firm. If a properly configured Claude setup saves each fee earner ninety minutes of production work per day, that is seven and a half hours per week redirected towards advisory work, client relationships, or simply handling more engagements without adding headcount. For a ten-person firm, that is 75 hours per week of reclaimed capacity. In a profession where capacity is the binding constraint during busy season, that number changes the economics of the practice.

But the more significant shift is not time saved. It is quality of output. The management reports become more insightful. The advisory letters are more thorough. The client correspondence is more consistent. The work that goes out under your firm’s name improves because senior people spend less time on production and more time on the thinking that clients actually value. For a firm trying to move up the value chain from compliance into advisory, this is the enabler.

The firms that treat AI as a compliance shortcut will save some time. The firms that treat it as an advisory capability multiplier will change what their practice is worth.

The Adoption Gap

Right now, most accountancy practices are in one of three positions. Some have ignored AI entirely. Some have let individual staff experiment in an unstructured way. A very small number have implemented it properly around how their practice actually works.

The gap between the third group and the first two is widening every month.

Anthropic, the company behind Claude, now holds 73% of the enterprise AI market, up from 40% just three months ago. Their platform is purpose-built for professional services: they do not train on your data, they offer SSO and centralised admin controls, and the security and data governance questions that legitimately concerned firms two years ago have been resolved. This is not experimental technology. It is the platform that large organisations across every sector are standardising on.

The barrier is not cost. It is not security. It is not the technology itself. The barrier is implementation. Someone needs to understand how your practice actually works, configure the tool around those workflows, build the Skills that encode your firm’s standards, and train your people so that adoption sticks beyond the first week.

The firms that buy seats and hope for the best get expensive novelty that fades within 90 days. The firms that implement properly build a structural advantage that compounds every month.

What Proper Implementation Looks Like

If you are a managing partner or practice leader reading this and thinking about making a move, here is what doing it properly actually involves.

First, map the real workflows. Not the process manual version. The actual daily reality of what your team produces, where time disappears, and which tasks would be transformed by having an intelligent system configured around your practice. Every firm is slightly different. A firm focused on owner-managed businesses has different workflows to one serving audit clients or property portfolios. The implementation needs to reflect how your firm actually operates.

Second, configure Claude around your firm. Load your house style, your standard templates, your preferred disclosures, your client intelligence, your technical guidance library, your best examples of management reports and advisory letters. A generic Claude account is useful. A Claude account loaded with your firm’s accumulated knowledge is transformative.

Third, build Skills for your highest-frequency tasks. Identify the five or ten tasks that consume the most hours across the practice. Build a Skill for each one. Management accounts commentary. Year-end review. Tax advisory letters. Client onboarding. Engagement letters. Board packs. Each Skill captures not just the process but your firm’s quality standards and voice.

Fourth, train the people individually. Group training is a starting point, but adoption lives or dies at the individual level. A tax partner needs a different configuration to an audit manager. A bookkeeping team member needs a different setup to a client relationship partner. The implementation has to reflect how each person actually works.

Fifth, sustain it. The firms that get transformative results are the ones that keep building. New Skills get created as confidence grows. The library expands into new use cases. An internal champion maintains and evolves the setup. This is not a one-off project. It is a new operating capability for the practice.

The Question That Matters

The question facing UK accountancy practices right now is not whether AI will change the profession. That question has been answered. The question is whether you implement it properly (configured around how your firm actually works, with the depth of setup that produces real results) or whether you buy seats, run a workshop, and end up in the same position six months from now.

I work with professional services firms to do the first version. A focused enablement sprint that configures Claude around your practice, builds the Skills your team needs, loads your content, and trains your people so that by the end of the engagement, every person has a working system they will actually use.

Not a product. Not a training day. Not a slide deck about what AI could theoretically do. A working implementation, configured to your firm, ready to use.